Allow use of custom providers for delegate agents.

hermes_cli/runtime_provider.py

@@ -18,6 +18,45 @@
 from hermes_constants import OPENROUTER_BASE_URL
 
 
+def _normalize_custom_provider_name(value: Optional[str]) -> str:
+    return (value or "").strip().lower().replace(" ", "-")
+
+
+def _resolve_custom_provider(requested: Optional[str]) -> Optional[Dict[str, Any]]:
+    requested_norm = _normalize_custom_provider_name(requested)
+    if not requested_norm or requested_norm == "auto":
+        return None
+
+    config = load_config()
+    custom_providers = config.get("custom_providers") or []
+    if not isinstance(custom_providers, list):
+        return None
+
+    for entry in custom_providers:
+        if not isinstance(entry, dict):
+            continue
+        name = entry.get("name")
+        base_url = (entry.get("base_url") or "").strip().rstrip("/")
+        if not name or not base_url:
+            continue
+
+        custom_name = _normalize_custom_provider_name(str(name))
+        if requested_norm not in {custom_name, f"custom:{custom_name}"}:
+            continue
+
+        api_mode = "codex_responses" if "chatgpt.com/backend-api/codex" in base_url.lower() else "chat_completions"
+        return {
+            "provider": custom_name,
+            "api_mode": api_mode,
+            "base_url": base_url,
+            "api_key": (entry.get("api_key") or "").strip(),
+            "source": "config.custom_providers",
+            "requested_provider": requested_norm,
+        }
+
+    return None
+
+
 def _get_model_config() -> Dict[str, Any]:
     config = load_config()
     model_cfg = config.get("model")
@@ -120,6 +159,10 @@ def resolve_runtime_provider(
     """Resolve runtime provider credentials for agent execution."""
     requested_provider = resolve_requested_provider(requested)
 
+    custom_runtime = _resolve_custom_provider(requested_provider)
+    if custom_runtime is not None:
+        return custom_runtime
+
     provider = resolve_provider(
         requested_provider,
         explicit_api_key=explicit_api_key,

tests/test_runtime_provider_resolution.py

@@ -200,7 +200,58 @@ def test_explicit_openrouter_skips_openai_base_url(monkeypatch):
     assert resolved["api_key"] == "or-test-key"
 
 
+def test_resolve_runtime_provider_custom_provider_from_config(monkeypatch):
+    monkeypatch.setattr(
+        rp,
+        "load_config",
+        lambda: {
+            "custom_providers": [
+                {
+                    "name": "local",
+                    "base_url": "http://0.0.0.0:8080/v1/",
+                    "api_key": "***",
+                    "model": "qwen-36b-q4_k_xl",
+                }
+            ]
+        },
+    )
+
+    resolved = rp.resolve_runtime_provider(requested="local")
+
+    assert resolved["provider"] == "local"
+    assert resolved["api_mode"] == "chat_completions"
+    assert resolved["base_url"] == "http://0.0.0.0:8080/v1"
+    assert resolved["api_key"] == "***"
+    assert resolved["requested_provider"] == "local"
+    assert resolved["source"] == "config.custom_providers"
+
+
+def test_resolve_runtime_provider_custom_provider_with_non_fqdn_hostnames(monkeypatch):
+    monkeypatch.setattr(
+        rp,
+        "load_config",
+        lambda: {
+            "custom_providers": [
+                {"name": "localhost", "base_url": "http://localhost:11434/v1/", "api_key": "", "model": "llama"},
+                {"name": "lab-box", "base_url": "http://lab-box:8080/v1/", "api_key": "token", "model": "qwen"},
+                {"name": "mesh", "base_url": "http://llm-node.local:9000/v1/", "api_key": "mesh-key", "model": "phi"},
+            ]
+        },
+    )
+
+    localhost = rp.resolve_runtime_provider(requested="localhost")
+    bare_host = rp.resolve_runtime_provider(requested="lab-box")
+    mdns = rp.resolve_runtime_provider(requested="mesh")
+
+    assert localhost["base_url"] == "http://localhost:11434/v1"
+    assert localhost["provider"] == "localhost"
+    assert bare_host["base_url"] == "http://lab-box:8080/v1"
+    assert bare_host["provider"] == "lab-box"
+    assert mdns["base_url"] == "http://llm-node.local:9000/v1"
+    assert mdns["provider"] == "mesh"
+
+
 def test_resolve_requested_provider_precedence(monkeypatch):
-    monkeypatch.setenv("HERMES_INFERENCE_PROVIDER", "nous")
+
     monkeypatch.setattr(rp, "_get_model_config", lambda: {"provider": "openai-codex"})
     assert rp.resolve_requested_provider("openrouter") == "openrouter"

tests/tools/test_delegate.py

@@ -94,6 +94,15 @@ def test_empty_input(self):
 
 
 class TestDelegateTask(unittest.TestCase):
+    def setUp(self):
+        # Keep these unit tests isolated from any real ~/.hermes delegation config
+        # on the machine running the suite.
+        self._load_config_patcher = patch("tools.delegate_tool._load_config", return_value={})
+        self._load_config_patcher.start()
+
+    def tearDown(self):
+        self._load_config_patcher.stop()
+
     def test_no_parent_agent(self):
         result = json.loads(delegate_task(goal="test"))
         self.assertIn("error", result)
@@ -316,6 +325,32 @@ def test_nous_provider_resolves_nous_credentials(self, mock_resolve):
         self.assertEqual(creds["api_key"], "nous-agent-key-xyz")
         mock_resolve.assert_called_once_with(requested="nous")
 
+    def test_custom_provider_resolves_from_config(self):
+        """Custom providers defined in config.yaml should be valid delegation targets."""
+        parent = _make_mock_parent(depth=0)
+        cfg = {"model": "zh-qwen27-thinking", "provider": "local"}
+
+        with patch(
+            "hermes_cli.runtime_provider.load_config",
+            lambda: {
+                "custom_providers": [
+                    {
+                        "name": "local",
+                        "base_url": "http://100.100.0.11:8931/v1/",
+                        "api_key": "no-api-key",
+                        "model": "zh-qwen27-thinking",
+                    }
+                ]
+            },
+        ):
+            creds = _resolve_delegation_credentials(cfg, parent)
+
+        self.assertEqual(creds["model"], "zh-qwen27-thinking")
+        self.assertEqual(creds["provider"], "local")
+        self.assertEqual(creds["base_url"], "http://100.100.0.11:8931/v1")
+        self.assertEqual(creds["api_key"], "no-api-key")
+        self.assertEqual(creds["api_mode"], "chat_completions")
+
     @patch("hermes_cli.runtime_provider.resolve_runtime_provider")
     def test_provider_resolution_failure_raises_valueerror(self, mock_resolve):
         """When provider resolution fails, ValueError is raised with helpful message."""